Human intention understanding in untrimmed videos aims to watch a natural video and predict what the person’s intention is. Currently, exploration of predicting human intentions in untrimmed videos is far from enough. On the one hand, untrimmed videos with mixed actions and backgrounds have a significant long-tail distribution with concept drift characteristics. On the other hand, most methods can only perceive instantaneous intentions, but cannot determine the evolution of intentions. To solve the above challenges, we propose a loss based on Instance Confidence and Class Accuracy (ICCA), which aims to alleviate the prediction bias caused by the long-tail distribution with concept drift characteristics in video streams. In addition, we propose an intention-oriented evolutionary learning method to determine the intention evolution pattern (from what action to what action) and the time of evolution (when the action evolves). We conducted extensive experiments on two untrimmed video datasets (THUMOS14 and ActivityNET v1.3), and our method has achieved excellent results compared to SOTA methods. The code and supplementary materials are available at https://github.com/Jennifer123www/UntrimmedVideo.

Network pruning is a figurative model compression technique designed to lighten and accelerate neural network models. Most existing pruning methods prioritize the selection of filters by their importance or apply regularization based on the properties of individual filters, neglecting the internal connections within combinations of multiple filters. This work introduces a pruning method termed Combinational Filter Deletion (CFD), which incorporates a straightforward yet effective evaluation metric based on the diversity of filter combination distributions to reveal the characteristics inherent to multiple filter interactions. CFD enables the exploration of an expanded search space, offering a greater array of choices and leveraging the intrinsic information of conventional layers. Moreover, this method is both general and non-exclusive, capable of enhancing the efficacy of other single-filter-based pruning techniques.